RNASEQ WITHOUT A REFERENCE

Transcription

1 RNASEQ WITHOUT A REFERENCE Experimental Design Assembly in Non-Model Organisms And other (hopefully useful) Stuff Meg Staton mstaton1@utk.edu University of Tennessee Knoxville, TN

2 I. Project Design

3 Things you need to know BEFORE you begin Cost Read Count Replicates Pro Tip: Who is your resident statistician? Buy them a coffee and make friends.

4 Replicates What? Biological Replicates independent biological sample, processed separately and barcoded Technical Replicates independent library construction or sequencing of the same biological sample Technical reproducibility is very good for RNASeq Biological variation is much greater! Different genes have different variances and are potentially subject to different errors and biases. Thinking About RNA Seq Experimental Design for Measuring Differential Gene Expression: The Basics Marioni, J.C., et al. (2008) RNA-seq: An assessment of technical reproducibility and

5 Replicates How many? beyond a depth of 10 million reads, replicates provide more statistical power than depth for detecting differential gene expression Many people say at least 3 this enables the t-test What if one fails? (Fishers exact test can utilize no replicates)

6 Replicates Software? Both EdgeR and DeSeq will calculate variance from replicates (but neither do a t-test) From the horse s mouth: to use something like a t test, you need enough replicates to estimate a variance for each gene. With two groups of five samples, you are already entering the regime there this should work well. For comparison, also try a tool that pools information from several genes to get better confidence in variance estimates, such as our DESeq or the Smyth group's edger. Of course, we like to claim that DESeq is better than edger, and for only two or three replicates, I do think so, but for five or more replicates, edger's "moderation" feature really pays off. So, even though I don't like admitting this, for your set-up [of 5 replicates per treatment], edger should work better than DESeq. -Simon Anders on SeqAnswers

7 Replicates And Blocks? Randomized Block Design Randomize - assigning individuals at random to treatments in an experiment Blocking - Experimental units are grouped into homogeneous clusters in an attempt to improve the comparison of treatments Example all organisms from the same location are blocks, multiple locations used Example - each block is a cultivar, with individuals from that cultivar randomly assigned to a treatment

8 Read Count - How to Decide? Standards, Guidelines and Best Practices for RNA-Seq V1.0 (June 2011) The ENCODE Consortium What are you trying to do? Compare two mrna samples for differential expression (30M PE per sample) Discover novel elements, perform more precise quantification, especially of lowly expressed transcripts ( M PE per sample) What resources do you already have? Well assembled and annotated genomes single ends, shorter reads De novo longer reads, paired ends What is being published in your community?

9 Read Count How to Decide? (cont.) Blogosphere disagrees Need half the coverage, double the replicates! Current experiments indicate that we are NOT discovering significantly more transcripts with a hiseq run vs a miseq run. (At least not transcripts that look like genes) A deep biased view

10 Scotty You need more power! Scotty is a web service to plan RNA-Seq experiments that measure differential gene expression. Prototype data required Pilot data -at least two replicates of either control or treatment Pre-loaded data

11 Scotty up to $20k User Inputs Used in the Analysis Control columns in pilot data: 3 Test columns in pilot data: 3 Cost per replicate, control: $200 Cost per replicate, test: $200 Cost per million reads: $23 Alignment Rate: 90% Maximum cost of experiment: $20000 Percentage of genes detected: 50 At p value cutoff: 0.01 For the following true fold change: 2 Maximum percentage of genes with low-powered (biased) measurements: 50 Least expensive: 6 replicates sequenced to a depth of 12 million reads aligned to genes per replicate. $5,712 Most powerful: 20 replicates sequenced to a depth of 34 million reads aligned to genes per replicate. $19,640

12 When is enough enough? Average coverage is not a helpful metric for RNASeq relative expression varies by orders of magnitude Saturation curves the number of transcript references with < 10 read alignments New Discovery Rate How many new genes are being discovered with each additional slice of data?

13 Reality: Variation in # of Sequences green ash blackgum white ash honeylocust sugar maple black walnut sweetgum white oak black cherry redbay - 2,000 4,000 6,000 8,000 10,000 12,000 14,000 16,000 18,000 Thousands

14 Spreading libraries across lanes?

15 Spreading libraries across lanes? Balanced Block Design

16 What s right for your experiment? Cost Read Count? Replicates

17 II. De novo transcriptome sequencing - assembly

18 Model Organism reference genome Trimmomatic quality and adapter trimming Tophat align reads to consensus genome HTSeq count read alignments by library DESeq differential expression

19 Non-model Organism NO reference genome Trimmomatic quality and adapter trimming Trimmomatic quality and adapter trimming Trinity - assembly Tophat align reads to consensus genome Bowtie2 Realign reads to consensus transcripts HTSeq count read alignments by library HTSeq Count read alignments by library DESeq differential expression DESeq differential expression

20 Problems with de novo assemblies plant species have larger and more complex genome sizes and structures than animal species tremendous diversity in both size and structure From a plant perspective Polyploidy Gene family proliferation Heterozygosity Repetitive element proliferation

21 Why plants are difficult (cont) Our best reference, Arabidopsis thaliana has a genome that underwent a 30% reduction in genome size and at least nine rearrangements in the short time since its divergence to Arabidopsis lyrata Maize pan genome - Intraspecific variations of as much as 38.8% from the average of 5.5 pg/2n nucleus driven by LTR retrotransposon expansion Conifer genome sizes Loblolly pine 22Gb (7x bigger than human) largest genome contains roughly 60,000,000,000 more base pairs than the smallest genome Often these difficulties make transcriptome sequencing more attractive than whole genome sequencing!

22 Hornett and Wheat. Quantitative RNA-Seq analysis in non-model species: assessing transcriptome assemblies as a scaffold and the utility of evolutionary divergent genomic reference species. BMC Genomics 2012, 13:361 Problems with de novo assemblies Results Highly fragmented assemblies Chimeras: Paralogs, alleles and alternative splicing variants mushed together or fragmented Metrics based upon contig lengths (e.g. mean, median, N50) do not provide quantitative insights into how much of the target species transcriptome is represented in the de novo TA.

23 Chestnut

24 De novo transcriptome assemblies Is there a close relative with a sequenced genome? How close is close enough? Align then assemble Assemble then align

25 Assemble then align First, assemble Next, align to a close relative Main Problems: Fragmented assemblies gene pieces are scattered in a different consensus pieces More difficult to sort out gene family members Main Advantages: Alignment to a close relative can identify exon/exon boundaries (sort out alternative splicing) Less bias can discover novel gene sequences Weber et al. Strategies for transcriptome analysis in nonmodel plants Am J Bot (2):

26 Green Ash (Fraxinus pennsylvanica) Trimmed reads Trimmed bases ozone project (miseq) 5,151, ,286,276 Tissues (miseq) 21,362,330 2,926,958,573 Tissues (hiseq) 442,863,286 42,122,511,244 Stress (miseq) 27,470,000 3,650,984,673 Stress (hiseq) 350,952,104 35,411,991,796 Data 847,799,220 84,862,732,562 Green Ash transcripts 107,611 peptides 52,899 % ORF discovery 49% 55 libraries Plus 41 technical replicates

27 Ash Genome Richard Buggs' lab Queen Mary, University of London (QMUL) British Ash Tree Genome Project Fraxinus excelsior 89,285 scaffolds, with an N50 of 99 kbp, and total size of 875 Mbp Green Ash 36,944 genes 36,893 proteins transcripts 107,611 peptides 52,899 % ORF discovery 49%

28 Ash Genome From perspective of the genome 36,893 proteins from genome 29,782 have a match to our RNASeq proteome (81%) From perspective of the transcriptome 52,899 proteins from RNASeq 47,657 have a match to the genome proteins (90%) 36,944 genes from genome 35,298 have a match to our RNASeq transcripts (96%) 107,611 transcripts from RNASeq 80,628 have a match to the genome transcripts (75%) BLAST 1e-10

29 Align then assemble First, map reads to (distant) reference Next, do local assemblies for each gene Main Problems Read alignment may be poor due to lack of sequence similarity Gene family expansion/contraction Main Advantage Transcript assembly is less likely to be fragmented Even where it is fragmented, you can identify all the fragments that originate from a single locus Weber et al. Strategies for transcriptome analysis in nonmodel plants Am J Bot (2):

30 Hornett and Wheat. Quantitative RNA-Seq analysis in non-model species: assessing transcriptome assemblies as a scaffold and the utility of evolutionary divergent genomic reference species. BMC Genomics 2012, 13:361 Assemble then align Reasonable Distant References Bad Worse *Likely to vary by phylogenetic neighborhood

31 De novo transcriptome assemblies Completely reference free What are they useful for? Transcriptome characterization Whats there? Resource Building Enabling proteomics experiments Candidate gene discovery Targeted sequencing Marker discovery/development Sequence parents of a cross SNP array (May want to consider genotyping by sequencing/restriction site associated DNA techniques instead)

32 De novo transcriptome assemblies What to do if you want/need differential expression data? Long reads Paired ends, possibly with different insert sizes Analyze gene families for differential expression instead of individual genes Alter the parameters of your assembler Merge at a lower level of heterozygosity 98% or 97% Utilize a closely related relative with a sequenced genome

33 Trinity strategy Inchworm assembles the RNAseq data into the unique sequences of transcripts, often generating full-length transcripts for a dominant isoform, but then reports just the unique portions of alternatively spliced transcripts.

34 Trinity strategy Chrysalis clusters the Inchworm contigs into clusters and constructs complete de Bruijn graphs for each cluster. Each cluster represents the full transcriptonal complexity for a given gene (or sets of genes that share sequences in common). Chrysalis then partitions the full read set among these disjoint graphs.

35 Trinity strategy Butterfly then processes the individual graphs in parallel, tracing the paths that reads and pairs of reads take within the graph, ultimately reporting fulllength transcripts for alternatively spliced isoforms, and teasing apart transcripts that corresponds to paralogous genes.

36 Trinity Videos!

37 Trinity output deciphering the naming An example Fasta entry for one of the transcripts is formatted like so: >c115_g5_i1 len=247 path=[31015: : ] Component a collection of contigs that are likely to be derived from alternative splice forms or closely related paralogs Gene best guess at an individual locus Isoform alternative splicing events and alleles

38 II. De novo transcriptome sequencing after assembly

39 Trinity TransDecoder and Coverage Maximizes length and likelihood score of ORF Optionally, looks for a putative peptide that has a match to a Pfam domain Full-length transcript analysis for model and nonmodel organisms using BLAST+ Perl script analyze_blastplus_tophit_c overage.pl

40 Functional Annotation InterProScan 329,311 annotations 45,893 transcripts have at least one annotation (87%) 234,546 GO term assignments 29,666 transcripts with go terms (56%) Software: 72,706 PANTHER 51,025 Pfam 41,391 Gene3d 39,965 SUPERFAMILY 27,246 TMHMM 26,189 ProSiteProfiles 20,835 PRINTS 20,078 SMART 8,267 Coils 5,780 TIGR-FAM 2,456 SignalP_EUK 1,224 PIRSF 825 HAMAP

41 Making data public NCBI Short Read Archive stores raw sequence data from "next-generation" sequencing technologies including 454, IonTorrent, Illumina, SOLiD, Helicos and Complete Genomics. In addition to raw sequence data, SRA now stores alignment information in the form of read placements on a reference sequence.

42 NCBI SRA SRA format includes data and metadata Convert using the SRA Toolkit (linux, mac and windows versions available)

43 Upload to SRA

44 NSF Hardwood Genomics Project